ScalarType diff(ublas::vector<ScalarType> const & v1, VCLVectorType const & v2)
{
   ublas::vector<ScalarType> v2_cpu(v2.size());
   viennacl::backend::finish();  //workaround for a bug in APP SDK 2.7 on Trinity APUs (with Catalyst 12.8)
   viennacl::copy(v2.begin(), v2.end(), v2_cpu.begin());

   for (unsigned int i=0;i<v1.size(); ++i)
   {
      if (v2_cpu[i] != v1[i])
        return 1;
   }

   return 0;
}
ScalarType diff(ublas::vector<ScalarType> const & v1, VCLVectorType const & v2)
{
   ublas::vector<ScalarType> v2_cpu(v2.size());
   viennacl::backend::finish();  //workaround for a bug in APP SDK 2.7 on Trinity APUs (with Catalyst 12.8)
   viennacl::copy(v2.begin(), v2.end(), v2_cpu.begin());

   for (unsigned int i=0;i<v1.size(); ++i)
   {
      if ( std::max( std::fabs(v2_cpu[i]), std::fabs(v1[i]) ) > 0 )
         v2_cpu[i] = std::fabs(v2_cpu[i] - v1[i]) / std::max( std::fabs(v2_cpu[i]), std::fabs(v1[i]) );
      else
         v2_cpu[i] = 0.0;
   }

   return norm_inf(v2_cpu);
}
ScalarType diff(std::vector<ScalarType> const & v1, VCLVectorType const & v2)
{
   std::vector<ScalarType> v2_cpu(v2.size());
   viennacl::backend::finish();  //workaround for a bug in APP SDK 2.7 on Trinity APUs (with Catalyst 12.8)
   viennacl::copy(v2.begin(), v2.end(), v2_cpu.begin());

   ScalarType norm_inf = 0;
   for (unsigned int i=0;i<v1.size(); ++i)
   {
     if ( std::max( std::fabs(v2_cpu[i]), std::fabs(v1[i]) ) > 0 )
     {
       ScalarType tmp = std::fabs(v2_cpu[i] - v1[i]) / std::max( std::fabs(v2_cpu[i]), std::fabs(v1[i]) );
       if (tmp > norm_inf)
         norm_inf = tmp;
     }
   }

   return norm_inf;
}